Sync GitHub main 6bb97fb9a95926f8118bd61bec366af28afa0e5c (part 3)

research/generated/CATALOG_AUDIT_REPORT.json

@@ -2,8 +2,8 @@
   "status": "clean",
   "audit": {
     "clean": false,
-    "question_reference_count": 5250,
-    "hypothesis_reference_count": 3911,
+    "question_reference_count": 5256,
+    "hypothesis_reference_count": 3918,
     "missing_questions": [
       "RQ-AI-DESIGN-001",
       "RQ-AI-EFF-001",

research/generated/CATALOG_AUDIT_REPORT.md

@@ -5,8 +5,8 @@
 The canonical ResearchRegistry remains authoritative. Historical/design references and test-only fixtures are listed with explicit reasons; an unknown missing identifier fails CI.
 
 ## Summary
-- Question references: 5250
-- Hypothesis references: 3911
+- Question references: 5256
+- Hypothesis references: 3918
 - Missing questions: 27
 - Missing hypotheses: 24
 - Registry link issues: 0

research/generated/CROSSWALK.md

@@ -37,7 +37,7 @@ All registered RQs are projected. Empty direction coordinates remain explicit un
 | RQ-CUDA-SCALE-001 | unresolved | unclassified | unresolved | unresolved | H-CUDA-SCALE-001-A, H-CUDA-SCALE-001-B | unresolved | unresolved | unresolved |
 | RQ-CUDA-STRUCT-001 | unresolved | unclassified | unresolved | unresolved | H-CUDA-STRUCT-001-A, H-CUDA-STRUCT-001-B | unresolved | unresolved | unresolved |
 | RQ-DET-001 | unresolved | unclassified | unresolved | III, IV, X, XI | H-SNN-003-A | EXP-2026-0001, EXP-BATCH-20260906200118-03, EXP-BATCH-20260908200906-03, EXP-BATCH-20260914074039-03, EXP-DET-0001, EXP-GEN-0023, EXP-GEN-0046 | EVID-2026-01, EVID-2026-03, EVID-2026-05, EVID-2026-07, EVID-2026-09, EVID-2026-11, EVID-2026-13, EVID-2026-15 | unresolved |
-| RQ-EMB-001 | unresolved | unclassified | unresolved | IV | H-EMB-001-A, H-EMB-001-B | EXP-BATCH-20260906200118-28, EXP-BATCH-20260909223705-28, EXP-BATCH-20260914074039-28, EXP-EMB-0001, EXP-EMB001B-MATCHED-V2-20260920, EXP-EMB001B-TEMPORAL-V3-20260920, EXP-EMP-20260910 | unresolved | unresolved |
+| RQ-EMB-001 | unresolved | unclassified | unresolved | IV | H-EMB-001-A, H-EMB-001-B | EXP-BATCH-20260906200118-28, EXP-BATCH-20260909223705-28, EXP-BATCH-20260914074039-28, EXP-EMB-0001, EXP-EMB001B-MATCHED-V2-20260920, EXP-EMB001B-TEMPORAL-V3-20260920, EXP-EMB001B-UNSEEN-V4-20261024, EXP-EMP-20260910 | unresolved | unresolved |
 | RQ-EMB-002 | unresolved | unclassified | unresolved | unresolved | H-EMB-002-A, H-EMB-002-B | EXP-BATCH-20260909223705-69, EXP-BATCH-20260914074039-69, EXP-EMB002A-DELAY-SWEEP-V4-20260924, EXP-EMB002A-PROPRIOCEPTION-V3-20260920, EXP-EMB002B-TRANSITION-V5-20260924, EXP-EMP-20260910 | unresolved | unresolved |
 | RQ-EMB-003 | unresolved | unclassified | unresolved | unresolved | H-EMB-003-A | unresolved | unresolved | unresolved |
 | RQ-EMB-004 | unresolved | unclassified | unresolved | unresolved | H-EMB-004-A | EXP-BATCH-20260909223705-71, EXP-BATCH-20260914074039-71, EXP-EMP-20260910 | unresolved | unresolved |

research/generated/DISSERTATION_MAP.md

@@ -38,7 +38,7 @@ Source: `research/publications/2026-09-17_recursive-epistemics_v1.8/parts/04_emp
 | --- | --- | --- | --- | --- | --- |
 | RQ-5D-005 | open | unresolved | unclassified | EXP-BATCH-20260906200118-38, EXP-BATCH-20260909223705-38, EXP-BATCH-20260914074039-38, EXP-EMP-20260910 | unresolved |
 | RQ-DET-001 | open | unresolved | unclassified | EXP-2026-0001, EXP-BATCH-20260906200118-03, EXP-BATCH-20260908200906-03, EXP-BATCH-20260914074039-03, EXP-DET-0001, EXP-GEN-0023, EXP-GEN-0046 | EVID-2026-01, EVID-2026-03, EVID-2026-05, EVID-2026-07, EVID-2026-09, EVID-2026-11, EVID-2026-13, EVID-2026-15 |
-| RQ-EMB-001 | open | unresolved | unclassified | EXP-BATCH-20260906200118-28, EXP-BATCH-20260909223705-28, EXP-BATCH-20260914074039-28, EXP-EMB-0001, EXP-EMB001B-MATCHED-V2-20260920, EXP-EMB001B-TEMPORAL-V3-20260920, EXP-EMP-20260910 | unresolved |
+| RQ-EMB-001 | open | unresolved | unclassified | EXP-BATCH-20260906200118-28, EXP-BATCH-20260909223705-28, EXP-BATCH-20260914074039-28, EXP-EMB-0001, EXP-EMB001B-MATCHED-V2-20260920, EXP-EMB001B-TEMPORAL-V3-20260920, EXP-EMB001B-UNSEEN-V4-20261024, EXP-EMP-20260910 | unresolved |
 | RQ-GW-002 | open | unresolved | unclassified | EXP-BATCH-20260909223705-79, EXP-BATCH-20260914074039-84 | unresolved |
 | RQ-GW-004 | open | unresolved | unclassified | EXP-BATCH-20260909223705-81, EXP-BATCH-20260914074039-86 | unresolved |
 | RQ-GW-006 | open | unresolved | unclassified | EXP-BATCH-20260909223705-83, EXP-BATCH-20260914074039-88 | unresolved |

research/generated/EVIDENCE_MATRIX.md

@@ -33,7 +33,7 @@ RQ-Status und Claim-Status sind unterschiedliche wissenschaftliche Zustände und
 | `RQ-SELF-002` | open | `H-SELF-002-A` | — | — | 0 | `EXP-BATCH-20260906200118-25`, `EXP-BATCH-20260908200906-25`, `EXP-BATCH-20260909223705-25`, `EXP-BATCH-20260914074039-25` | — | offen |
 | `RQ-STRUCT-001` | open | `H-STRUCT-001-A` | — | — | 0 | `EXP-BATCH-20260906200118-26`, `EXP-BATCH-20260908200906-26`, `EXP-BATCH-20260909223705-26`, `EXP-BATCH-20260914074039-26` | — | offen |
 | `RQ-MEM-001` | open | `H-MEM-001-A` | — | — | 0 | `EXP-BATCH-20260906200118-27`, `EXP-BATCH-20260909223705-27`, `EXP-BATCH-20260914074039-27` | — | offen |
-| `RQ-EMB-001` | open | `H-EMB-001-A`, `H-EMB-001-B` | — | — | 0 | `EXP-BATCH-20260906200118-28`, `EXP-BATCH-20260909223705-28`, `EXP-BATCH-20260914074039-28`, `EXP-EMB-0001`, `EXP-EMB001B-MATCHED-V2-20260920`, `EXP-EMB001B-TEMPORAL-V3-20260920`, `EXP-EMP-20260910` | — | offen |
+| `RQ-EMB-001` | open | `H-EMB-001-A`, `H-EMB-001-B` | — | — | 0 | `EXP-BATCH-20260906200118-28`, `EXP-BATCH-20260909223705-28`, `EXP-BATCH-20260914074039-28`, `EXP-EMB-0001`, `EXP-EMB001B-MATCHED-V2-20260920`, `EXP-EMB001B-TEMPORAL-V3-20260920`, `EXP-EMB001B-UNSEEN-V4-20261024`, `EXP-EMP-20260910` | — | offen |
 | `RQ-LLM-001` | open | `H-LLM-001-A` | — | — | 0 | `EXP-BATCH-20260906200118-29`, `EXP-BATCH-20260909223705-29`, `EXP-BATCH-20260914074039-29` | — | offen |
 | `RQ-ETH-001` | open | `H-ETH-001-A`, `H-ETH-001-B`, `H-ETH-001-C`, `H-ETH-001-D`, `H-ETH-001-E` | — | — | 0 | `EXP-BATCH-20260906200118-30`, `EXP-BATCH-20260909223705-30`, `EXP-BATCH-20260914074039-30` | — | offen |
 | `RQ-ETH-002` | open | `H-ETH-002-A` | — | — | 0 | `EXP-BATCH-20260906200118-31`, `EXP-BATCH-20260909223705-31`, `EXP-BATCH-20260914074039-31` | — | offen |

research/generated/META_RESEARCH_CROSSWALK.json

@@ -1058,6 +1058,7 @@
         "EXP-EMB-0001",
         "EXP-EMB001B-MATCHED-V2-20260920",
         "EXP-EMB001B-TEMPORAL-V3-20260920",
+        "EXP-EMB001B-UNSEEN-V4-20261024",
         "EXP-EMP-20260910"
       ],
       "hypotheses": [
@@ -5570,6 +5571,15 @@
         "research/experiments/EXP-EMB001B-TEMPORAL-V3-20260920/review_request.json.review.json"
       ]
     },
+    {
+      "authority": "review_record_only_not_EVID_or_replication",
+      "experiment_id": "EXP-EMB001B-UNSEEN-V4-20261024",
+      "manifest_review_status": "PENDING",
+      "reconciliation_state": "recorded_review_present_check_status_binding",
+      "review_artifacts": [
+        "research/experiments/EXP-EMB001B-UNSEEN-V4-20261024/review_request.json"
+      ]
+    },
     {
       "authority": "review_record_only_not_EVID_or_replication",
       "experiment_id": "EXP-EMB002A-DELAY-SWEEP-V4-20260924",

research/generated/REVIEW_QUEUE.md

@@ -233,6 +233,7 @@ This projection reconciles manifest review flags with append-only review artefac
 | EXP-EMB-0001 | unknown | unresolved | review_not_found_in_experiment_root |
 | EXP-EMB001B-MATCHED-V2-20260920 | PENDING | research/experiments/EXP-EMB001B-MATCHED-V2-20260920/review_request.json, research/experiments/EXP-EMB001B-MATCHED-V2-20260920/review_request.json.review.json | recorded_review_present_check_status_binding |
 | EXP-EMB001B-TEMPORAL-V3-20260920 | PENDING | research/experiments/EXP-EMB001B-TEMPORAL-V3-20260920/review_request.json, research/experiments/EXP-EMB001B-TEMPORAL-V3-20260920/review_request.json.review.json | recorded_review_present_check_status_binding |
+| EXP-EMB001B-UNSEEN-V4-20261024 | PENDING | research/experiments/EXP-EMB001B-UNSEEN-V4-20261024/review_request.json | recorded_review_present_check_status_binding |
 | EXP-EMB002A-DELAY-SWEEP-V4-20260924 | PENDING | research/experiments/EXP-EMB002A-DELAY-SWEEP-V4-20260924/review_request.json, research/experiments/EXP-EMB002A-DELAY-SWEEP-V4-20260924/review_request.json.human-review.json, research/experiments/EXP-EMB002A-DELAY-SWEEP-V4-20260924/review_request.json.review.json | recorded_review_present_check_status_binding |
 | EXP-EMB002A-PROPRIOCEPTION-V3-20260920 | PENDING | research/experiments/EXP-EMB002A-PROPRIOCEPTION-V3-20260920/review_request.json, research/experiments/EXP-EMB002A-PROPRIOCEPTION-V3-20260920/review_request.json.review.json | recorded_review_present_check_status_binding |
 | EXP-EMB002B-TRANSITION-V5-20260924 | PENDING | research/experiments/EXP-EMB002B-TRANSITION-V5-20260924/review_request.json, research/experiments/EXP-EMB002B-TRANSITION-V5-20260924/review_request.json.review.json | recorded_review_present_check_status_binding |

research/generated/STAGE_MATRIX.md

@@ -6,14 +6,14 @@ Stage 0-10 is a maturity/development axis, not the research-direction taxonomy.
 | --- | --- | --- | --- |
 | 0 | Einzelne Nervenzelle | 0.925 | Die eingefrorene Stage-0-Konformitätsstudie stützt den numerischen Modellvertrag für Izhikevich-2003 und lif-current-v1 im deklarierten Operating Envelope. Das ist Modell-/Softwarekonformität, keine biologische Gleichwertigkeit, keine allgemeine Langzeittrajektorien-Identität und keine unabhängige Replikation. |
 | 1 | Kleines SNN | 0.85 | Stage 1 besitzt jetzt zwei getrennte kanonische, human-reviewte EVID-Linien: Topologie EVID-2026-19 im 64-Neuronen/246-Kanten-Regime und Temporal Order EVID-2026-20 im sechsneuronigen Zweikanal-Task. Der Temporal-Human-Review dokumentiert Perfect-Score-Sättigung und die simultane Bedingung als nicht-inferentiellen Task-Adequacy-Control. Das begründet weder Kognition, Lernen/Gedächtnis, Skalierbarkeit, biologische Äquivalenz, 5D-Überlegenheit noch unabhängige Replikation. |
-| 2 | Stabiles rekurrentes SNN | 0.65 | Stabile deterministische Runs stützen Stabilität im geprüften Bereich; sie sind kein Nachweis von Gedächtnis oder Kognition. |
-| 3 | Plastisches Nervengewebe | 0.55 | STDP, Drei-Faktor-Lernen, Homöostase und Strukturplastizität sind implementierte Mechanismen; ihre wissenschaftliche Wirksamkeit bleibt interventions- und EVID-gebunden. |
-| 4 | Spezialisierte neuronale Areale | 0.55 | Modality-specific pathways und Gateway-Verträge sind technisch vorhanden; E01-E05 bleiben DATA bis zur unabhängigen wissenschaftlichen Prüfung. |
-| 5 | Integriertes künstliches Nervensystem | 0.55 | Der geschlossene Regelkreis ist in einer synthetischen deterministischen Umgebung belegt; reale Embodiment- oder biologische Interozeptionsclaims sind nicht gedeckt. |
-| 6 | Gedächtnis und Weltmodell | 0.4 | Working/episodic foundations und Ein-Schritt-Vorhersage sind keine Semantization, kein hierarchisches Predictive Coding und kein generatives Weltmodell. |
-| 7 | Selbstmodell und verkörperte Identität | 0.28 | Versionierte technische Identität und Behavior State sind Grundlagen, aber kein kausales Selbstmodell und kein Bewusstseinsnachweis. |
-| 8 | Autonome lebenslange Entwicklung | 0.23 | Interferenz-Screens und Forschungsplanung sind keine Evidenz für autonomes lebenslanges Lernen. |
-| 9 | Hochintegrierte künstliche Kognition | 0.15 | Architektur- und Experimententwürfe sind keine beobachtete integrierte Kognition. |
-| 10 | Bewusstseinsforschung | 0.13 | Diese Stufe ist ausschließlich eine Forschungs- und Governance-Frontier. Kein Score begründet Bewusstsein, Sentienz oder moralischen Status. |
+| 2 | Stabiles rekurrentes SNN | 0.75 | Stabile deterministische Runs stützen Stabilität im geprüften Bereich; sie sind kein Nachweis von Gedächtnis oder Kognition. |
+| 3 | Plastisches Nervengewebe | 0.65 | STDP, Drei-Faktor-Lernen, Homöostase und Strukturplastizität sind implementierte Mechanismen; ihre wissenschaftliche Wirksamkeit bleibt interventions- und EVID-gebunden. |
+| 4 | Spezialisierte neuronale Areale | 0.65 | Modality-specific pathways und Gateway-Verträge sind technisch vorhanden; E01-E05 bleiben DATA bis zur unabhängigen wissenschaftlichen Prüfung. |
+| 5 | Integriertes künstliches Nervensystem | 0.65 | Der geschlossene Regelkreis ist in einer synthetischen deterministischen Umgebung belegt; reale Embodiment- oder biologische Interozeptionsclaims sind nicht gedeckt. |
+| 6 | Gedächtnis und Weltmodell | 0.45 | Working/episodic foundations und Ein-Schritt-Vorhersage sind keine Semantization, kein hierarchisches Predictive Coding und kein generatives Weltmodell. |
+| 7 | Selbstmodell und verkörperte Identität | 0.325 | Versionierte technische Identität und Behavior State sind Grundlagen, aber kein kausales Selbstmodell und kein Bewusstseinsnachweis. |
+| 8 | Autonome lebenslange Entwicklung | 0.4 | Interferenz-Screens und Forschungsplanung sind keine Evidenz für autonomes lebenslanges Lernen. |
+| 9 | Hochintegrierte künstliche Kognition | 0.3 | Architektur- und Experimententwürfe sind keine beobachtete integrierte Kognition. |
+| 10 | Bewusstseinsforschung | 0.3 | Diese Stufe ist ausschließlich eine Forschungs- und Governance-Frontier. Kein Score begründet Bewusstsein, Sentienz oder moralischen Status. |
 
 > Authority boundary: implementation != DATA != reviewed EVID != independent replication; Human Review != independent replication; DOI != peer review. This projection executes and promotes nothing.

research/generated/verification/determinism_infrastructure.json

@@ -1,12 +1,12 @@
 {
   "schema_version": 1,
   "status": "verified",
-  "test_run_head": "893ff761d34c3997715f4e526a010f683dc98b78",
-  "tested_commit": "893ff761d34c3997715f4e526a010f683dc98b78",
-  "tested_tree_digest": "989452dad584e04fe75c850798324e442c139994c7bcd4532416e75bb2d20037",
+  "test_run_head": "359b17b7759beadf883354ba61953e51c3a134f5",
+  "tested_commit": "359b17b7759beadf883354ba61953e51c3a134f5",
+  "tested_tree_digest": "fd40f85877572792bc3df30279ce3f40c19cdfe32e3171bc030708d561d0a112",
   "scope": "restore_determinism",
   "scope_digest": "2948f92769ea97075e943e55f06fd99318869379e84a207c88fa5de50b468a23",
-  "timestamp": "2026-10-24T10:39:56+0000",
+  "timestamp": "2026-10-24T15:42:12+0000",
   "python": "3.13.16",
   "test_files": [
     "tests/test_rng_persistence.py",

research/generated/verification/phase_b_gate_status.json

@@ -3,23 +3,23 @@
     "artifact": "tests/test_baseline.json",
     "collection_errors": 0,
     "failed": 0,
-    "passed": 2100,
+    "passed": 2113,
     "python": "3.13.16",
-    "skipped": 52
+    "skipped": 54
   },
   "continuous_integration": {
     "conclusion": "success",
-    "head_sha": "893ff761d34c3997715f4e526a010f683dc98b78",
-    "run_id": 40086567103,
-    "workflow": "Verification Refresh #191"
+    "head_sha": "359b17b7759beadf883354ba61953e51c3a134f5",
+    "run_id": 40107936098,
+    "workflow": "Verification Refresh #197"
   },
   "final_validation": {
     "command": "python scripts/generate_baseline.py",
     "failed": 0,
-    "passed": 2100,
-    "skipped": 52
+    "passed": 2113,
+    "skipped": 54
   },
-  "generated": "2026-10-24T10:39:59.996027+00:00",
+  "generated": "2026-10-24T15:42:15.504162+00:00",
   "limitations": [
     "Scientific evidence status remains independent from release engineering readiness.",
     "External Hugging Face proxy smoke remains non-blocking operational follow-up."
@@ -42,8 +42,8 @@
     "overall": "passed"
   },
   "source_freeze": {
-    "commit": "893ff761d34c3997715f4e526a010f683dc98b78",
-    "tested_tree_digest": "989452dad584e04fe75c850798324e442c139994c7bcd4532416e75bb2d20037"
+    "commit": "359b17b7759beadf883354ba61953e51c3a134f5",
+    "tested_tree_digest": "fd40f85877572792bc3df30279ce3f40c19cdfe32e3171bc030708d561d0a112"
   },
   "status": "verified",
   "verification_artifacts": [

research/generated/verification/restore_determinism.json

@@ -2,19 +2,19 @@
   "schema_version": 1,
   "suite": "restore_determinism",
   "status": "verified",
-  "test_run_head": "893ff761d34c3997715f4e526a010f683dc98b78",
-  "tested_commit": "893ff761d34c3997715f4e526a010f683dc98b78",
-  "tested_tree_digest": "989452dad584e04fe75c850798324e442c139994c7bcd4532416e75bb2d20037",
+  "test_run_head": "359b17b7759beadf883354ba61953e51c3a134f5",
+  "tested_commit": "359b17b7759beadf883354ba61953e51c3a134f5",
+  "tested_tree_digest": "fd40f85877572792bc3df30279ce3f40c19cdfe32e3171bc030708d561d0a112",
   "scope": "restore_determinism",
   "scope_digest": "2948f92769ea97075e943e55f06fd99318869379e84a207c88fa5de50b468a23",
-  "timestamp": "2026-10-24T10:39:06+0000",
+  "timestamp": "2026-10-24T15:41:29+0000",
   "python": "3.13.16 (main, Oct  1 2026, 02:37:54) [GCC 13.3.0]",
   "K": 500,
   "N": 1000,
   "seed": 42,
   "config_sha256": "b9dbcba2952663b8f75c839bcf935e66539f35c7ddcc1294f4157f869aae42d4",
-  "pid_C1": 12230,
-  "pid_C2": 12234,
+  "pid_C1": 12306,
+  "pid_C2": 12310,
   "digest_A": "eca7a8b2b81bc9bf3045933bd12113a31204d01355636bfd0e4432f11bed77b7",
   "digest_B": "eca7a8b2b81bc9bf3045933bd12113a31204d01355636bfd0e4432f11bed77b7",
   "digest_C": "eca7a8b2b81bc9bf3045933bd12113a31204d01355636bfd0e4432f11bed77b7",

research/generated/verification/single_listener.json

@@ -2,13 +2,13 @@
   "schema_version": 1,
   "suite": "single_listener",
   "status": "verified",
-  "timestamp": "2026-10-24T10:39:28.033603",
+  "timestamp": "2026-10-24T15:41:49.461963",
   "python_version": "3.13.16",
-  "test_run_head": "893ff761d34c3997715f4e526a010f683dc98b78",
-  "tested_commit": "893ff761d34c3997715f4e526a010f683dc98b78",
-  "tested_tree_digest": "989452dad584e04fe75c850798324e442c139994c7bcd4532416e75bb2d20037",
+  "test_run_head": "359b17b7759beadf883354ba61953e51c3a134f5",
+  "tested_commit": "359b17b7759beadf883354ba61953e51c3a134f5",
+  "tested_tree_digest": "fd40f85877572792bc3df30279ce3f40c19cdfe32e3171bc030708d561d0a112",
   "scope": "runtime_integration",
-  "scope_digest": "ae1c0af61b4b289b1dbe7662fd0542084168240da0e1d06e295b544dd3f80480",
+  "scope_digest": "6301a4441fb62c46afa4437da0f92a3d987e654189cdf1f03210c508de3ab411",
   "test_command": "python -m pytest tests/test_single_listener.py -q",
   "proofs": {
     "port_initially_free_or_explicitly_rejected": true,

research/generated/verification/structural_e2e.json

@@ -2,11 +2,11 @@
   "schema_version": 1,
   "suite": "structural_e2e",
   "status": "verified",
-  "timestamp": "2026-10-24T10:39:41.572513",
+  "timestamp": "2026-10-24T15:42:00.694612",
   "python_version": "3.13.16",
-  "test_run_head": "893ff761d34c3997715f4e526a010f683dc98b78",
-  "tested_commit": "893ff761d34c3997715f4e526a010f683dc98b78",
-  "tested_tree_digest": "989452dad584e04fe75c850798324e442c139994c7bcd4532416e75bb2d20037",
+  "test_run_head": "359b17b7759beadf883354ba61953e51c3a134f5",
+  "tested_commit": "359b17b7759beadf883354ba61953e51c3a134f5",
+  "tested_tree_digest": "fd40f85877572792bc3df30279ce3f40c19cdfe32e3171bc030708d561d0a112",
   "scope": "structural_e2e",
   "scope_digest": "e325bba5d44e1c067c045d545544915040cf5c38def8fb19fbf02d3f23abc522",
   "test_command": "python -m pytest tests/test_structural_e2e.py -q",

research/generated/verification/structural_live_loop.json

@@ -2,11 +2,11 @@
   "schema_version": 1,
   "suite": "structural_live_loop",
   "status": "verified",
-  "timestamp": "2026-10-24T10:39:42.751153",
+  "timestamp": "2026-10-24T15:42:01.661071",
   "python_version": "3.13.16",
-  "test_run_head": "893ff761d34c3997715f4e526a010f683dc98b78",
-  "tested_commit": "893ff761d34c3997715f4e526a010f683dc98b78",
-  "tested_tree_digest": "989452dad584e04fe75c850798324e442c139994c7bcd4532416e75bb2d20037",
+  "test_run_head": "359b17b7759beadf883354ba61953e51c3a134f5",
+  "tested_commit": "359b17b7759beadf883354ba61953e51c3a134f5",
+  "tested_tree_digest": "fd40f85877572792bc3df30279ce3f40c19cdfe32e3171bc030708d561d0a112",
   "scope": "structural_live_loop",
   "scope_digest": "31f26a0c2ec56eec09444556c4e7814e5b56aa54ae444930c17aee3c599c8475",
   "test_command": "python -m pytest tests/test_structural_live_loop.py -q",

research/preregistrations/PREREG-5D-MATCHED-R2.freeze.json

@@ -0,0 +1,18 @@
+{
+  "status": "FROZEN_DESIGN_ONLY",
+  "date": "2026-10-24",
+  "execution_authorized": false,
+  "evaluation_executed": false,
+  "base_commit": "fdfd0568",
+  "hashes": {
+    "research/preregistrations/PREREG-5D-MATCHED-R2.md": "041543ec8103f810e2d7eab7c1867ea79b487b163e7a4e53a8a31303f93f6f54",
+    "scripts/run_stage1_topology_v2.py": "144de33384814838c6f33ecbba3ff80e840882f77396e25e90f67dfd7b23972e",
+    "scripts/run_stage1_topology_v3_r1.py": "0a7dced0f158eae315a72a33e2de9cfc69b6a99d92eb0755b74e95903f0513df",
+    "configs/learning_experiment.yaml": "6f6cd457caf6216c286fee47ef03b529a0b496ed2dd53ee96b85d3b51e7b3a1c"
+  },
+  "pending": [
+    "dedicated R2 runner and analysis enforcement",
+    "synthetic preflight and seed freshness",
+    "execution authorization"
+  ]
+}

research/preregistrations/PREREG-5D-MATCHED-R2.md

@@ -0,0 +1,60 @@
+# Matched 5D ablation R2 — design freeze
+
+Date: 2026-10-24. **FROZEN_DESIGN_ONLY; execution_authorized=false.**
+Binding: `RQ-5D-005 / H-5D-005-A`. This prospective design complements, and
+does not overwrite, `PREREG-5D-005` or historical Stage-1 studies. It does
+not unblock the failed recruitment-matching gate of efficiency R1.
+
+## Frozen design
+
+Six arms: 1d=(64,1,1,1,1), 2d=(8,8,1,1,1), 3d=(4,4,4,1,1),
+5d=(4,2,2,2,2), 5d_shuffled, random_graph. Every arm has 64 neurons and
+246 directed acyclic edges: each source i has min(4,63-i) outgoing edges.
+Coordinates, normalized distances, deterministic seed-based tie breaking,
+coordinate shuffle and random-target sampling use `all_coords`,
+`canonical_coords`, `normalized_distance`, `edge_list` and `build_network`
+from `scripts/run_stage1_topology_v2.py`, bound by hash in the design freeze.
+Record realized edge lists and degree sequences. Edge count and out-degree
+are matched; in-degree, path lengths and graph geometry may differ and are
+the intervention, not hidden matching assumptions. This is a comparison of
+specific graph construction rules, not a pure effect of abstract dimension.
+
+All synapses weight 55, delay 1 tick; input labels 0–3, output labels 60–63;
+one 100-current impulse at tick 0; 128 evaluation ticks; identical initial
+neuron parameters from `configs/learning_experiment.yaml`. No learning,
+structural plasticity, optimizer, warm-up training or hyperparameter search:
+all have exactly zero updates/trials. Weight 55 was chosen historically;
+it is not retuned against these new outcomes. Seeds 96001–96030, paired across
+arms, each condition reconstructed independently. No calibration on these
+seeds. Equal training/search effort means zero in every arm, not unknown.
+
+Primary outcome: sum of cumulative active-neuron fractions for the first 32
+ticks, using the time-resolved observation convention in
+`run_stage1_topology_v3_r1.py` (range 0–32, not divided by window length).
+Primary contrast: 5d minus 3d. Fixed directional claim of benefit requires a
+positive paired mean and one-sided exact sign-test p<0.05 (ties excluded,
+effective n reported). Thirty complete pairs mandatory; zero non-tied pairs
+give p=1. No secondary contrast may rescue the primary. Report a 95% paired
+bootstrap interval (20,000 resamples, RNG seed 20261024) descriptively.
+All other arms, half-activation latency, first-output latency and total spikes
+are secondary/descriptive. This endpoint measures early activation, not task
+accuracy, intelligence or energy efficiency. A faster graph need not be better.
+
+Resource ceilings: one CPU worker, float64 scalar simulation; no GPU;
+2 GiB peak process RSS; 60 seconds/arm/seed; 3 CPU hours for all 180 runs;
+2 GiB saved output. Record wall time/RSS at one-second intervals. Stop and
+retain partial data on exceeded resources, NaN, invalid graph, count mismatch
+or runtime failure. No replacement seeds, automatic reruns or parameter
+adjustment. Invalid/incomplete runs imply NOT_TESTED, not no 5D effect.
+
+Before evaluation a dedicated R2 runner must enforce this design, run only
+synthetic integrity fixtures, verify evaluation-seed freshness, record all
+source/config/environment hashes and obtain execution authorization. The
+bound historical V3 runner must NOT be invoked directly: its hard-coded
+seed plan and multiple-primary-endpoint analysis differ from this R2 plan.
+Design freeze therefore does not claim executable freeze or completed DATA.
+
+The freeze record binds this document and source dependencies before any R2
+evaluation. Changes require a versioned amendment and new hashes. A technical
+or scientific review may reject this design; freezing does not certify its
+adequacy, power or independence.

research/preregistrations/PREREG-EMB001B-UNSEEN-V4.json

@@ -0,0 +1,24 @@
+{
+  "preregistration_id": "PREREG-EMB001B-UNSEEN-V4",
+  "research_question": "RQ-EMB-001",
+  "hypothesis": "H-EMB-001-B",
+  "mode": "PROSPECTIVE_EXPLORATORY_CAUSAL_DIAGNOSTIC",
+  "execution_authorized": true,
+  "authorization_basis": "User requested completion of the controlled closed-loop/yoked comparison in this chat on 2026-10-24; DATA only, not human scientific review.",
+  "scientific_evidence": false,
+  "human_review_required": true,
+  "automatic_evidence_promotion": false,
+  "seeds": [94001, 94002, 94003, 94004, 94005, 94006, 94007, 94008, 94009, 94010, 94011, 94012, 94013, 94014, 94015, 94016, 94017, 94018, 94019, 94020],
+  "ticks": 1000,
+  "conditions": ["donor_unperturbed", "yoked_unperturbed", "closed_challenge", "yoked_challenge", "feedback_absent_challenge", "pd_challenge"],
+  "disturbance": "Even seed: +0.003 Nm; odd seed: -0.003 Nm, ticks [250,500); zero otherwise. Identical schedule in all four challenge arms. Donor and yoked-unperturbed arms receive zero torque.",
+  "causal_intervention": "Record same-seed unperturbed closed-loop donor first. Freeze its sensor tape before any challenge run. Challenge yoked controller consumes this tape without access to the challenged body's sensors. Closed loop receives current challenged-body sensors. No shifted tape, different targets, learning or condition-specific tuning.",
+  "primary_outcome": "tracking RMSE over post-step body positions in ticks [250,1000), target +0.25 before tick 500 and -0.25 thereafter",
+  "primary_contrast": "yoked_challenge minus closed_challenge, paired by seed; positive means closed-loop advantage",
+  "secondary": ["full-window tracking RMSE", "feedback_absent_challenge minus closed_challenge", "pd_challenge minus closed_challenge", "sensor-path RMS", "motor command tapes"],
+  "analysis": "Report every paired difference, mean, median, min, max and positive/zero/negative counts, separately by disturbance sign and pooled. No p-values, no confirmatory threshold, no causal attribution to learned representations.",
+  "mandatory_checks": ["clean committed source before execution", "120 complete runs of 1000 ticks", "no runtime errors", "same initial neural/body state and graph within seed", "unperturbed yoked trace equals donor trace", "challenged yoked observed sensor tape equals frozen donor", "challenged yoked motor tape equals donor despite body disturbance", "all challenge arms receive identical disturbance schedule", "feedback-absent observed sensors equal zero", "nonzero disturbance changes yoked physical trajectory"],
+  "budgets": {"learning_updates": 0, "hyperparameter_trials": 0, "neurons": 6, "cpu_workers": 1, "maximum_seconds": 600, "maximum_output_bytes": 100000000},
+  "failure_rule": "Retain partial outputs and failures. No replacement seeds, automatic reruns or post-result tuning. Integrity failure means INVALID, not hypothesis refutation.",
+  "claim_boundary": "Fixed engineered error receptors and fixed spike-to-torque decoder in a synthetic one-joint world. This tests access to challenge-contingent feedback, not biological equivalence, independent replication, learned body schema, general intelligence or 5D merit. PD arm is attribution reference, not matched neural mechanism."
+}

research/preregistrations/PREREG-MEM-COMPRESSION-R1.md

@@ -0,0 +1,149 @@
+# OBJ-MEM-COMPRESSION-001 — R1 review-ready protocol
+
+Date: 2026-10-24. Status: **REVIEW_READY_NOT_FROZEN**. Execution authorized:
+**false**. No evaluation data collected. This document resolves design choices;
+it does not attest to a reviewed or executable implementation.
+
+Canonical RQ/H binding remains pending for `OBJ-MEM-COMPRESSION-001`.
+No unregistered identifier is substituted for the existing memory
+hypotheses. R1 approval and registry admission remain necessary. The authority
+for this gate is edition 1.8, section 19.7: source/config hashes, canonical
+RQ/H binding and R1 human approval must be bound before FROZEN.
+
+## Question and fixed decision
+
+Can semantic prototype replay use at most 10% of the full raw-replay memory
+budget while preserving at least 95% of its retention? This is a narrow
+compression claim. Five percent relative loss is the maximum acceptable
+loss, not five accuracy percentage points.
+
+After a technically valid complete run, retain SemanticMemory as a candidate
+compression component only if the simultaneous non-inferiority and byte
+budget gates below pass. Additionally require non-inferiority to matched-byte
+raw replay and a positive semantic advantage over random prototypes.
+Otherwise de-center SemanticMemory in the default architecture for this task;
+retain its implementation as an experimental option. An inconclusive interval
+does not prove inferiority, but does not justify retaining a central role.
+Missing or invalid data mean NOT_TESTED, not a negative scientific finding.
+Changing the task, loss tolerance or denominator requires a new registration.
+
+## Task, seeds, learning and controls
+
+Use Split-MNIST tasks (0,1), (2,3), (4,5), (6,7), (8,9), in that order;
+1,000 training and 200 held-out test examples per class. Select the first
+eligible examples after a per-class permutation from the paired seed. Store
+the exact dataset file hashes and selected indices before evaluation. The
+test split never updates memory, normalization, readout, selection or tuning.
+Evaluation seeds are 95001–95030; implementation-only fixture seeds 31–33
+must use synthetic inputs, not MNIST evaluation outcomes.
+
+Use the existing sparse encoder (pooling 4; thresholds .2,.4,.6,.8),
+OnlineSoftmaxReadout learning rate .025, same initial readout per seed,
+10,000 update slots (2,000/task); later tasks use 400 replay slots replacing
+current-example updates. The replay slot schedule and all test exposure are
+identical across replay conditions. No hyperparameter search (zero trials),
+no condition-specific training, no extra replay updates. Prototype extraction
+may observe each arriving training example once; it cannot revisit discarded
+episodes. Charge its computational cost separately, never hide it as free
+preprocessing. Freeze the encoder/readout/selection implementation before use.
+
+| Arm | Retained budget | Purpose |
+| --- | --- | --- |
+| raw_full | B100 | Full raw-replay reference |
+| semantic_10 | floor(B100/10) | Compression candidate |
+| raw_10 | floor(B100/10) | Essential matched-byte replay baseline |
+| random_prototype_10 | floor(B100/10) | Representation-content control |
+| no_replay | 0 memory bytes | Same 10,000 slots, current examples only |
+| learning_off | 0 memory bytes | Initial readout, no optimizer updates; diagnostic only |
+
+Raw selection uses a seeded reservoir of past training records, never test
+performance. Random prototypes match label/task and feature count of the
+semantic candidates but sample feature IDs uniformly without access to the
+semantic IDs. Memory capacity is bytes, not object count. Underfilled arms
+retain unused capacity; they do not receive extra training or information.
+The learning-off arm is not presented as a matched-update efficacy baseline.
+
+## Complete 10% accounting contract
+
+Define B100 at each task boundary as the canonical serialized size of every
+raw record seen so far, including features, label, task ID, length and index
+metadata, plus its replay-selection state. Define B10=floor(B100/10).
+The same B10 applies to all three restricted arms, independently of their
+actual compression success. Report boundary and peak values, not just final
+snapshot size. If header/metadata alone exceeds B10, declare capacity failure.
+
+Serialization is uncompressed, little-endian, schema-versioned: header and
+all counts use uint64; labels/task IDs uint32; sorted unique feature IDs
+uint32 with uint64 lengths; numeric prototype accumulators float64; integer
+support counters uint64. Include all record headers, offsets, membership
+lists, provenance IDs needed at runtime, RNG state, lookup indices, retained
+episodic buffer, support histograms, pending consolidation state, caches,
+eviction metadata and replay cursors. No hidden full raw store, duplicate
+representation, external index, pickle compression or disk spill is allowed.
+Report a per-category ledger after every update and consolidation operation.
+The state serializer must prove that restoring only the charged state
+reproduces the next replay selection and update.
+
+The primary budget is replay-memory information, not process RSS. Immutable
+encoder parameters and program code are excluded only when byte-identical
+across replay arms. The common readout is excluded only with identical
+architecture, dtype, allocated bytes and update-slot count; its learned values
+may differ. Record its entire state and size separately. Extra trainable
+parameters, optimizer buffers or learned encoders are forbidden. One current
+input is excluded with identical shape/dtype and no retained copy. Training
+data are streamed through a coordinator inaccessible
+to arm memory after the current update. Test data stay outside all learners.
+Temporary workspaces are capped at B10 additional bytes in restricted arms,
+B100 in raw_full; publish peak workspace and combined peak. Python allocator
+overhead, shared library memory and whole-process RSS are measured separately.
+No claim of tenfold total-RAM reduction may follow from this retained-state
+ratio. Offline output logs are write-only to the learner and cannot act as
+memory. Any persistent arm-specific auxiliary state belongs in the ledger.
+
+When capacity would be exceeded, use deterministic eviction before the next
+update: lowest support, then oldest creation index, then canonical record ID
+for prototypes; seeded reservoir order for raw replay. Charge temporary
+incoming/consolidation records to the workspace ceiling. Never prune based
+on test outcomes. The executable serializer, eviction and peak-accounting
+tests are mandatory prerequisites; the existing object-count matching from
+CL-003 does not satisfy this contract.
+
+## Outcomes, uncertainty and failure handling
+
+After each task evaluate each learned task. Retention R(seed,arm) is the
+unweighted mean final accuracy on tasks 1–4 (800 test examples/task are NOT
+assumed: exactly 400/task). Task 5 is excluded from retention and reported
+separately as plasticity. Also report taskwise accuracy and forgetting.
+Primary non-inferiority contrast per paired seed:
+`R(semantic_10) - 0.95 * R(raw_full)`.
+Require its lower 95% paired-bootstrap percentile bound >=0, 20,000 resamples,
+analysis RNG seed 20261024. Report ratio of means as descriptive only;
+if raw_full mean <=0, ratio is undefined and the study is invalid.
+
+Two additional conjunctive gates: lower 95% bound of
+`R(semantic_10)-0.95*R(raw_10)` >=0; lower 95% bound of
+`R(semantic_10)-R(random_prototype_10)` >0. All three gates must pass;
+this intersection rule does not select whichever comparison succeeds.
+Also require raw_full mean retention >=0.50 and positive lower 95% bound
+against no_replay to establish a functioning replay reference. Thresholds
+are prospective engineering choices, not established literature effects.
+Thirty seeds is a fixed feasibility choice, not a guarantee of power.
+
+Run all 30 paired seeds, six arms; CPU only, one worker, 2 GiB process RSS,
+30 minutes/seed/arm, 90 CPU hours total maximum, 20 GiB output ceiling.
+Measure wall time and RSS at least once per second. Exceeding any cap stops
+the study with partial artifacts retained. No replacement seeds, retries,
+post-hoc exclusions or outcome-dependent stopping. Numerical errors, budget
+violations, dataset drift, leaks and missing pairs block positive conclusions.
+
+## Freeze checklist still required
+
+1. Register the proposed RQ/H and obtain R1 human review of this exact text.
+2. Implement and test byte serializer, restore/selection equivalence, streaming
+   isolation, deterministic eviction, full-ledger peak checks and analysis.
+3. Bind clean Git commit/tree, config, dataset/index, source and test hashes,
+   environment, seed freshness scan and the review record in a freeze artifact.
+4. Obtain separate execution authorization after the complete freeze.
+
+Neither the earlier approval of the research idea nor this chat's request to
+prepare the protocol is represented as a scientific R1 review of this new plan.

research/preregistrations/PREREG-S1-TOPO-REFERENCE-R2.freeze.json

@@ -0,0 +1,107 @@
+{
+  "authorization_rule": {
+    "all_required": [
+      "canonical R2 freeze record exists and all bound hashes match",
+      "R2 runner independence/blinding scan passes at the exact freeze commit",
+      "R2 seed-freshness gate passes at the exact freeze commit",
+      "R2 immutable dynamics-reuse validation passes with source hashes matching the historical R1 parity artifacts",
+      "R2 topology-mapping parity passes for all registered conditions",
+      "no R1 or R2 reference evaluation DATA exist before authorization",
+      "verifier decision-rule contract tests match the frozen R2 preregistration",
+      "an explicit separate human execution-authorization record names the frozen commit and runner hash"
+    ],
+    "automatic_authorization_forbidden": true,
+    "current_value": false,
+    "human_authority": "Thomas Heisig or a separately documented authorized human reviewer"
+  },
+  "canonical_target_digest": "f3616151a7ab529c610f983f4fe855b84228505d806d516c502014a576d6d3b6",
+  "contract_tests": {
+    "tests\\test_stage1_reference_r2_contract.py": "ff153bea7840ca9df7dae046f1bf64f473afa9ad88b29da4c87aca7775c977ce",
+    "tests\\test_stage1_reference_r2_preflight.py": "10f9ca088bcdc23a9f573107fbccca56a871c91ed62f76aff3f9c2d7c0da1bc3"
+  },
+  "decision_rule_digest": "2eca23d888f4846488a2a2db53b1702120580283c5d5a73dae26667c854a01b0",
+  "dynamics_reuse_validation": {
+    "historical_artifacts_modified": false,
+    "pass": true,
+    "path": "research\\calibrations\\CAL-S1-TOPO-REFERENCE-DYNAMICS-REUSE-R2\\result.json",
+    "sha256": "73c6269f0a6886a77d9d824295d30979387a64c78a553127f7807c71af78c4b8"
+  },
+  "evaluation_seeds": [
+    820000001,
+    820000002,
+    820000003,
+    820000004,
+    820000005,
+    820000006,
+    820000007,
+    820000008,
+    820000009,
+    820000010,
+    820000011,
+    820000012,
+    820000013,
+    820000014,
+    820000015,
+    820000016,
+    820000017,
+    820000018,
+    820000019,
+    820000020
+  ],
+  "execution_authorized": false,
+  "freeze_commit": "e5d2fe71bd755b2ace7f12c80ba4c4624913a610",
+  "freeze_date": "2026-10-24",
+  "integrator_parity": {
+    "pass": true,
+    "path": "research\\calibrations\\CAL-S1-TOPO-REFERENCE-INTEGRATOR-R1\\result.json",
+    "reused_from_r1_with_source_hash_revalidation": true,
+    "sha256": "4779cdd5e4c8993bdc4f24a30a2d6c9206cd57f7606c107aa8dc2ea13388ea05"
+  },
+  "mechanism_audit": {
+    "pass": true,
+    "path": "research\\audits\\STAGE1_TOPOLOGY_REFERENCE_MECHANISM_AUDIT_20260927.json",
+    "sha256": "2269d45cd8d40bbcf585c8f1de2c20072273ffbcd5fe2cc028dc8a540912eef4"
+  },
+  "note": "R2 corrects the pre-DATA R1 mapping ambiguity. Freeze fixes the scientific contract only; separate explicit human execution authorization remains mandatory.",
+  "preregistration_id": "PREREG-S1-TOPO-REFERENCE-R2",
+  "preregistration_sha256_before_status_update": "4fbe9f36c7d7d97b5d83edef060e53ba05d70c85c571709c2c472be9ae35062c",
+  "reference_data_created": false,
+  "reference_runner": {
+    "execution_authorized": false,
+    "path": "reference\\stage1_topology_brian2\\runner_r2.py",
+    "sha256": "b4a7945210b4d836d5d9940c1695b92b6b0af8f5296f88ac8b3894974136f932"
+  },
+  "reference_verifier": {
+    "path": "scripts\\verify_stage1_topology_reference_r2.py",
+    "sha256": "be031bc8dfec4fd7b8c62e1ca86022d52beb6c31aba2dfc0b8867afc22ee1ef6"
+  },
+  "replication_credit_awarded": false,
+  "reset_parity": {
+    "pass": true,
+    "path": "research\\calibrations\\CAL-S1-TOPO-REFERENCE-RESET-R1\\result.json",
+    "reused_from_r1_with_source_hash_revalidation": true,
+    "sha256": "c1c50e07345e940b0cfe0599cabbbbd8b2b6f0447841fa989673ddfce522b247"
+  },
+  "sanitized_runner_protocol": {
+    "path": "reference\\stage1_topology_brian2\\reference_protocol_r2.json",
+    "sha256": "9a855c266cd9bf865f017c0d6e128735213eef6d8b2b2c508d6c64e153eccfc2"
+  },
+  "schema_version": 1,
+  "scientific_evidence": false,
+  "supersedes": "PREREG-S1-TOPO-REFERENCE-R1",
+  "synapse_delay_parity": {
+    "pass": true,
+    "path": "research\\calibrations\\CAL-S1-TOPO-REFERENCE-SYNAPSE-R1\\result.json",
+    "reused_from_r1_with_source_hash_revalidation": true,
+    "sha256": "42b051286007b315c5400871d893883d34be2847d5f165b08a70a91d0580907e"
+  },
+  "topology_mapping_parity": {
+    "pass": true,
+    "path": "research\\calibrations\\CAL-S1-TOPO-REFERENCE-TOPOLOGY-MAP-R2\\result.json",
+    "sha256": "ca96b049c57d8c75c5e1eb7112565756bbfb1eca7df402143c63dd8ab05e85dc"
+  },
+  "translation_specification": {
+    "path": "reference\\stage1_topology_brian2\\TRANSLATION_R2.md",
+    "sha256": "74caf0b06c5a36065a9f8b20dadfc2418185bd2254785307f12f9a3400bc1be6"
+  }
+}

research/preregistrations/PREREG-S1-TOPO-REFERENCE-R2.json

@@ -7,7 +7,7 @@
   "claim_id": "CLAIM-S1-TOPO-001",
   "protocol": "stage1_topology_reference_replication_r2",
   "mode": "PROSPECTIVE_CROSS_IMPLEMENTATION_PARTIAL_REPLICATION",
-  "status": "DRAFT_PRE_FREEZE_GATES_PENDING",
+  "status": "FROZEN_BEFORE_REFERENCE_EXECUTION_AUTHORIZATION",
   "execution_authorized": false,
   "scientific_evidence": false,
   "automatic_evidence_promotion": false,
@@ -241,7 +241,7 @@
       820000019,
       820000020
     ],
-    "seed_selection_rationale": "The originally drafted block 1200021-1200040 failed the executable freshness gate because seed 1200035 already existed in historical Batch artifacts. Before any reference DATA and before freeze, the entire block was discarded. The replacement block 810000001-810000020 was selected as a new contiguous bookkeeping block and returned no repository-search matches at selection time; the executable freshness gate must still pass at the exact freeze commit.",
+    "seed_selection_rationale": "The originally drafted block 1200021-1200040 failed the executable freshness gate because seed 1200035 already existed in historical Batch artifacts. Before any reference DATA and before freeze, the entire block was discarded. The replacement block 810000001-810000020 was selected as a new contiguous bookkeeping block and returned no repository-search matches at selection time; R2 subsequently uses the distinct block recorded in evaluation.seeds and reference_protocol_r2.json (820000001-820000020); the executable freshness gate must still pass at the exact freeze commit.",
     "conditions": [
       "1d",
       "2d",
@@ -476,7 +476,7 @@
     },
     "seed_freshness": {
       "required": true,
-      "rule": "Evaluation seeds 810000001-810000020 must be rechecked automatically at the freeze commit; any occurrence outside the preregistration and blinded reference_protocol.json blocks freeze.",
+      "rule": "The current R2 evaluation seeds in evaluation.seeds (820000001-820000020) must be rechecked automatically at the freeze commit; any occurrence outside the paths explicitly allowed by check_stage1_reference_seed_freshness_r2.py blocks freeze.",
       "status": "PENDING_R2"
     },
     "reset_parity": {
@@ -512,8 +512,8 @@
     "classification_consequence": "Even if a separate framework/codebase is used, this project-side implementation is PARTIAL_REPLICATION only. It cannot be relabeled EXTERNAL_REPLICATION without genuinely external authorship/control and a separately registered protocol."
   },
   "freeze_authorization": {
-    "allowed": false,
-    "rule": "R2 may freeze only after all source-bound dynamics gates, R2 independence/seed freshness, and exact topology-mapping parity pass.",
+    "allowed": true,
+    "rule": "All R2 pre-freeze gates passed and are hash-bound. This does not authorize reference evaluation.",
     "reference_runner_implementation_allowed": true,
     "reference_runner_execution_allowed": false
   },
@@ -526,8 +526,8 @@
       "replication_credit_awarded": false
     }
   ],
-  "freeze_record": "research/preregistrations/PREREG-S1-TOPO-REFERENCE-R2.freeze.json",
-  "freeze_semantics": "R2 is not frozen yet. Its corrected topology mapping and all comparison criteria must be frozen before any reference evaluation DATA. R1 remains historical and may never authorize execution.",
+  "freeze_record": "research\\preregistrations\\PREREG-S1-TOPO-REFERENCE-R2.freeze.json",
+  "freeze_semantics": "R2 scientific contract is frozen. Reference DATA remain forbidden until a separate explicit human execution-authorization record binds this freeze commit and runner hash.",
   "supersedes": "PREREG-S1-TOPO-REFERENCE-R1",
   "supersession": {
     "date": "2026-09-27",